Interference management in spectrally and energy efficient wireless networks

Interference Management with Limited CSI Sparse Spectrum Sensing in CRNs D2D Communications M2M Communications

Interference Management in Spectrallyand Energy Efficient Wireless Networks

Mohamed Seif, BSc

Wireless Intelligent Networks Center (WINC), Nile University, Egypt

August 10, 2016

Thesis Committee:

Prof. Mohamed NafieProf. Amr Elkeyi

Prof. Karim G. SeddikMohamed Seif, BSc Nile University

Interference Management in Spectrally and Energy Efficient Wireless Networks 1


Data Transmission

Storage

Energy

Mohamed Seif, BSc Nile University



1 Interference Management with Limited CSI

2 Sparse Spectrum Sensing in CRNs

3 D2D Communications

4 M2M Communications











The Big Problem in Wireless Communications

Figure: An illustrative example for a heterogeneous network.

Interference is a fundamental bottleneck in many wirelesssystemsInterference management is getting convoluted

Homogeneous → Heterogeneous

How to manage interference in an efficient manner?






Interference is a fundamental bottleneck in many wirelesssystems

Interference management is getting convolutedHomogeneous → Heterogeneous





















Coding Against Interference

Interference shaping using CSIT 1 is a key enabler for mitigatinginterference

1Channel state information at transmitter.




Coding Against Interference

Interference shaping using CSIT 1 is a key enabler for mitigatinginterference

1Channel state information at transmitter.




Closed Loop Systems

Figure: Illustration of the CSIT feedback and sharing process.




Challenges in Obtaining Global and Accurate CSIT

Tx

Channel Feedback

User1

User2

User3

Possible error sources in the CSI feedback processChannel estimation errorQuantization error (e.g., Compressed channel feedback)Feedback delay (Maddah Ali et al.’12)

CSIT sharing via backhaul links (e.g., CoMP in LTE)





Tx

Channel Feedback

User1

User2

User3


CSIT sharing via backhaul links (e.g., CoMP in LTE)





Tx

Channel Feedback

User1

User2

User3


CSIT sharing via backhaul links (e.g., CoMP in LTE)Mohamed Seif, BSc Nile University



Degrees of Freedom (DoF)

DoF notion:

1 Lizhong and Tse in IEEE IT Trans. 20032 Rigorous approximation to the network capacity in the high

SNR regime.

Mathematically,

C∑(P) = DoF log(P) + o(log(P)) (1)

where limP→∞o(log(P))

log(P) = 0.

Alternatively,It represents the number of interference-free signallingdimensions in the network.





DoF notion:1 Lizhong and Tse in IEEE IT Trans. 2003

2 Rigorous approximation to the network capacity in the highSNR regime.

Mathematically,



log(P) = 0.






DoF notion:1 Lizhong and Tse in IEEE IT Trans. 20032 Rigorous approximation to the network capacity in the high

SNR regime.

Mathematically,



log(P) = 0.





System Model

The received signal at the i th receiveris given by

Yi(t) = Hi(t)X(t) +Ni(t), i = 1, . . . ,K(2)

The total DoF of the network is definedas

DΣ(K ) = max(d1,d2,...,dK )∈D

d1 + d2 + ⋅ ⋅ ⋅ + dK

(3)

UE3

UE1 UE2

UE4

UEi

)(tHi

K-antenna Tx

UEK

Figure: Network Model




CSI Model

Perfect and global CSIR.

Three states of the availability of CSITabout each receiver:

Perfect CSIT (P): instantaneousand without error.Delayed CSIT (D): delay greaterthan or equal one time slotduration (coherence time) andwithout error.No CSIT (N): not available totransmitter at all.

UE3

UE1 UE2

UE4

UEi

)(tHi

K-antenna Tx

UEK

Introduced by Tandon and Shamai in IEEE IT Trans. 2012 for the 2-user BC




Alternating CSIT Model

The fraction of time associated withCSIT state S,

λS = ∑nt=1∑

Ki=1 I(Si(t) = S)

nK(4)

where n is the number of channeluses,

∑S∈{P,D,N}

λS = 1. (5)

UE3

UE1 UE2

UE4

UEi

)(tHi

K-antenna Tx

UEK




ICR SchemePhase I: Interference Creation

UE2

UE1

Tx

UE3

1u

2u

3u

),,( 321

1

1 uuuL

),,( 321

1

2 uuuI

),,( 321

1

3 uuuI

N

D

D

Figure: ICR scheme t = 1.





UE2

UE1

Tx

UE3

1v

2v

3v

),,( 321

1

1 vvvI

),,( 321

1

2 vvvL

),,( 321

1

3 vvvI

N

D

D






UE2

UE1

Tx

UE3

1p

2p

3p

),,( 321

1

1 pppI

),,( 321

1

2 pppI

),,( 321

1

3 pppL

D

D

N





ICR SchemePhase II: Interference Resurrection (Based on orthogonal projection pre-coding and PNC)

UE2

UE1

Tx

UE3

),,( 321

2

1 uuuL

),,( 321

2

2 vvvL

),,( 321

2

3 pppL

Old interference terms from UE3

P

N

P


Based on orthogonal projection pre-coding and PNCMohamed Seif, BSc Nile University



ICR SchemePhase II: Interference Resurrection (Based on orthogonal projection pre-coding and PNC)

UE2

UE1

Tx

UE3

),,( 321

3

1 uuuL

),,( 321

3

2 vvvL

),,( 321

3

3 pppL

Old interference terms from UE2

N

P

P


Based on orthogonal projection pre-coding and PNCMohamed Seif, BSc Nile University



ICR Scheme

UE2

UE1

Tx

UE3

1u2u 3u

1v2v 3v

1p2p 3p

Figure: D∑ = 95 , S5

123 = {NDD,DND,DDN,PPN,PNP}.




Synergistic Alternating CSIT

Phase I: Creation Phase II: Resurrection

(NDD,DND,DDN) (PPN,PNP)(NDD,DDN,DND) (PNP,PPN)(DND,DDN,DDN) (PPN,NPP)(DND,DDN,NDD) (NPP,PPN)(DDN,DND,NDD) (NPP,PNP)(DDN,NDD,DND) (PNP,NPP)

Table: All synergistic CSIT patterns for the 3-user BC.

Synergy Definition

Synergy is the interaction of multiple elements in a system toproduce an effect greater than the sum of their individualeffects.








Synergy Definition

Consider: S5123 = (NNN,DDD,DDD,DDD,PPP).

D∑(3) = 1 × 3

15 +1811 ×

915 + 3 × 3

15 = 9855 < 9

5








Synergy Definition

Consider: S5123 = (NNN,DDD,DDD,DDD,PPP).

D∑(3) = 1 × 3

15 +1811 ×

915 + 3 × 3

15 = 9855 < 9

5




Upper Bound on the DoF for the K-user BC

Bounds were introduced by Tandon et al.’13

DΣ(K ) = d1 + d2 + ⋅ ⋅ ⋅ + dK ≤K 2 + (K − 1)∑K

i=1 γi

2K − 1(6)

where,

γi =∑n

t=1 I(Si(t) = P)n

≤ γ,∀i = 1, . . . ,K (7)




DoF Region Characterization for the 3-user BC

Given perfect CSIT distribution (γ1,γ2,γ3),

P1: maxd1,d2,d3

d1 + d2 + d3

s.t. 3d1 + d2 + d3 ≤ 3 + 2γ1 (8)d1 + 3d2 + d3 ≤ 3 + 2γ2 (9)d1 + d2 + 3d3 ≤ 3 + 2γ3 (10)0 ≤ di ≤ 1, ∀i = 1,2,3 (11)

Closed form solution,

d∗

i =3 + 4γi −∑3

j=1,j≠i γj

5, ∀i = 1,2,3 (12)






P1: maxd1,d2,d3

d1 + d2 + d3


Closed form solution,

d∗

i =3 + 4γi −∑3

j=1,j≠i γj

5, ∀i = 1,2,3 (12)






P1: maxd1,d2,d3

d1 + d2 + d3


Solution,Given: (γ1, γ2, γ3) = (2

5 ,15 ,

15)

Optimal DoF tuple: d∗ = (0.84,0.64,0.64)

Achievable DoF tuple: d = (0.6,0.6,0.6).Conjecture: The outer bound can be achieved by addingmulti-cast messaging in the network.






P1: maxd1,d2,d3

d1 + d2 + d3



5 ,15 ,

15)

Optimal DoF tuple: d∗ = (0.84,0.64,0.64)Achievable DoF tuple: d = (0.6,0.6,0.6).

Conjecture: The outer bound can be achieved by addingmulti-cast messaging in the network.






P1: maxd1,d2,d3

d1 + d2 + d3



5 ,15 ,

15)

Optimal DoF tuple: d∗ = (0.84,0.64,0.64)Achievable DoF tuple: d = (0.6,0.6,0.6).

Conjecture: The outer bound can be achieved by addingmulti-cast messaging in the network.




ICR Scheme vs MAT Scheme

Achievable DoF for this network is given by

DΣ(K ) = K 2

2K − 1> K

1 + 12 + ⋅ ⋅ ⋅ +

1K

´¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¸¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¶Delayed CSIT - MAT scheme

(17)

and the distribution of fraction of time of the different states{P,D,N} required for our proposed scheme is

λP = (K − 1)2

2K 2 −K, λD = K − 1

2K − 1, λN = 1

K. (18)




Results

1 2 3 4 5 6 7 8 9 101

1.5

2

2.5

3

3.5

4

4.5

5

5.5

K (users)

DoF

sum

(K)

CSIT with alternationCSIT with all delayed

Figure: DoF comparison for broadcast channel between all delayedand alternating CSIT models.




Results

1 2 3 4 5 6 7 8 9 101

2

3

4

5

6

7

8

9

10

K (users)

DΣ(K

)

Upper bound on the K−user BC, γ=1Upper bound on alternating CSIT for the K−user BCAchievable DoF based on ICR scheme

Figure: DoF comparison for the K-user BC.











Sampling Theory

Shannon/Nyquist sampling theorem:

No information loss if we sampleat 2x signal bandwidthStorage/processing problem

Solution?

Yes, Compressive Sensing/Sampling




Sampling Theory



Solution?





Sampling Theory



Solution?





Compressive Sensing

Pioneered by E. Candes, T.Tao and D. DonohoSignal acquisition and compression in one stepSparsity in a certain transform domain (e.g., frequencydomain)




Compressive Sensing

Pioneered by E. Candes, T.Tao and D. Donoho

Signal acquisition and compression in one stepSparsity in a certain transform domain (e.g., frequencydomain)




Compressive Sensing

Pioneered by E. Candes, T.Tao and D. DonohoSignal acquisition and compression in one step

Sparsity in a certain transform domain (e.g., frequencydomain)




Compressive Sensing

Pioneered by E. Candes, T.Tao and D. DonohoSignal acquisition and compression in one stepSparsity in a certain transform domain (e.g., frequencydomain)




Compressive Sensing Formulation





RIP Condition:

(1 − δ) ∥x∥22 ≤ ∥Φx∥2

2 ≤ (1 + δ) ∥x∥22 . (19)





RIP Condition:

(1 − δ) ∥x∥22 ≤ ∥Φx∥2

2 ≤ (1 + δ) ∥x∥22 . (19)





Figure: Random measurements by φ (Gaussian).

Signal Recovery (`1 norm recovery):

minx∈RN

∥x∥1 s.t.∥y − φx∥2 ≤ ε (20)





Figure: Random measurements by φ (Gaussian).

Signal Recovery (`1 norm recovery):

minx∈RN

∥x∥1 s.t.∥y − φx∥2 ≤ ε (20)




CS for Spectrum Sensing

frequencyN channel sub-bands

Empty sub-band Occupied sub-band

Sparsity in PU occupation





frequencyN channel sub-bands

Empty sub-band Occupied sub-band

Sparsity in PU occupation





CR3

CR1 CR2

CR4

CRi

Fusion Center

Figure: Fusion based CRN.

Decision making: Majority-Rule, AND-Rule




CS for Spectrum Sensing in CRNs

Secondary network:

G(M,E)

Adjacency matrix A(k) ∈ RM×M :

aij(k) =⎧⎪⎪⎨⎪⎪⎩

1 if τij(k) >= τ, i ≠ j0 otherwise

(21)

aij modeled as a Bernoulli R.V. with prob.of success p

CR3

CR1 CR2

CR4

CRi

Figure: Infrastructure-lessCRN.




CS for Spectrum Sensing in CRNs

1 `1 norm recovery

2 Vector Consensus algorithm

bj(k) = ( 1M

(b(0) + 1Kp

K−1

∑t=0

B(t)aTj (t)))

(22)

Convergence will be achieved

limk→∞

bj(k) = b∗ (23)

Majority-Rule asymptotic behavior

limK→∞

Pd(K ) =N

∑j=1

M

∑i=⌈ M

2 ⌉

(Mi )(1−π11)M−iπi

11

(24)

CR3

CR1 CR2

CR4

CRi

Figure: Infrastructure-lessCRN.




Simulation Parameters

Parameter Symbol RealizationNo. channels N 200No. measurements T 30No. PU nodes P 4No. SU nodes M 12Minimum Distance dmin 10 (m)Area A 1000 (m) ×1000(m)Pathloss Exponent α 2




Results

0 5 10 15 20 250.9

0.95

1

SNR (dB)

Pd

0 5 10 15 20 250

2

4

6

8x 10

−3

SNR (dB)

Pfa

Centralized − Majority RuleInfrasturcture−less, K=20Infrasturcture−less, K=10Infrasturcture−less, K=1000

Centralized − Majority RuleInfrasturcture−less, K=20Infrasturcture−less, K=10Infrasturcture−less, K=1000

Figure: Performance comparison




Results

0 5 10 15 20 250.8

0.82

0.84

0.86

0.88

0.9

0.92

0.94

0.96

0.98

1

SNR (dB)

Pd

Centralized− Majority RuleInfrastructure−less, p=1Infrastructure−less, p=0.8Infrastructure−less, p=0.3Infrastructure−less, p=0.1

Figure: Effect of link quality




Results

0 5 10 15 20 250.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

SNR (dB)

Pd

Centralized − Majority Rule, T=50Infrasturcture−less, T=50Infrasturcture−less, T=40Infrasturcture−less, T=30Infrasturcture−less, T=20

Figure: Effect of number of measurements




Results

1 2 3 4 5 6 7 8 9 100.7

0.75

0.8

0.85

0.9

0.95

1

k (iterations)

Pd(k

)

Good connectivity, p=0.8, SNR=10 dBPoor connectivity, p=0.3, SNR =10 dBGood connectivity, p=0.8, SNR =5 dBPoor connectivity, p=0.3, SNR =5 dB

Figure: The convergence of consensus algorithm in terms probability ofdetection











Motivation

eNB

Figure: Traditional Cellular Network.

Applications are hungry!

Multimedia services

Existing infrastructure




Motivation

eNB



Multimedia services





Motivation

eNB



Multimedia services





Motivation

eNB



Multimedia services





Motivation

CUE

CUE

CUE

eNB

CUE

D2D Pair

D2D Pair

D2D Pair

Figure: D2D Communications.

Advantages:Offloading the cellular system → high data ratesReliable communications/Instant communicationsProximity effect → power saving




Motivation

CUE

CUE

CUE

eNB

CUE

D2D Pair

D2D Pair

D2D Pair


Advantages:Offloading the cellular system → high data rates

Reliable communications/Instant communicationsProximity effect → power saving




Motivation

CUE

CUE

CUE

eNB

CUE

D2D Pair

D2D Pair

D2D Pair


Advantages:Offloading the cellular system → high data ratesReliable communications/Instant communications

Proximity effect → power saving




Motivation

CUE

CUE

CUE

eNB

CUE

D2D Pair

D2D Pair

D2D Pair


Advantages:Offloading the cellular system → high data ratesReliable communications/Instant communicationsProximity effect → power saving




System Model

BS

CUE

D1

D2

Figure: Network Model: Cellular network with D2D network (shadedarea).




Cooperative Scheme

BS

CUE

D1

D2

Figure: Cooperative System, t = 1.




Cooperative Scheme

BS

CUE

D1

D2


xCTD1

=√αPCT

D1xC +

√(1 − α)PCT

D1xD2 (25)




Cooperative Scheme

BS

CUE

D1

D2


xCTD1

=√αPCT

D1xC +

√(1 − α)PCT

D1xD2 (25)




Problem Formulation

P1: maxρ,α

RCTC

s.t. PCTD1

≤ PT,max(EH constraint) (26)

RB,D1 ≥ RCTC (Decoding at D1) (27)

RCTD2

≥ RD2(Target rate for D2D pair) (28)ρ,α ∈ [0,1]. (29)




Simulation Parameters

Table: List of symbols.

Symbol Description Value

PB BS TX power 41 dBmN0 Noise power −100 dBmL Pathloss Exponent 1.8 − 3.8

dB,D1 Distance between B and D1 50 − 500 mdD1,C Distance between D1 and C 10 − 20 mdD1,D2 Distance between D1 and D2 5 − 20 mdB,C Distance between B and C 200 − 1000 m

dD1,D2 Distance between D1 and D2 5 − 20 mR Cell radius 500 m




Results

0 )1(log2

121,2

CT

DDR

2DR

UB

Cooperative Transmission(CT)

Direct Transmission(DT)

Figure: α vs RD2 .




Results

Pathloss Exponent2 2.2 2.4 2.6 2.8 3 3.2 3.4 3.6 3.8

RC

8

9

10

11

12

13

14

15

16

Without cooperationWith cooperation

Figure: RC vs Pathloss Exponent: PT,max = 29 dBm.




Results

dD1,D2(max)20 25 30 35 40 45 50 55 60 65 70

Pro

b. o

f suc

c. c

ance

latio

n

0.2

0.3

0.4

0.5

0.6

0.7

0.8

CUED2D-Rx

Figure: Probability of SIC vs dD1,D2(max) at D2











Motivation




System Model

Central Aggregator

1 t

2 t

3 t

K t

RN : Receive Antennas ix :Sparse signal of activity

Traffic Nature:1 Low data rate2 Sporadic → Sparse




Proposed Solutions

Figure: QPSK Constellation with threshold contour..

Detect ActivityDecode the data from the modulation alphabet A (e.g.,QPSK modulation)




Proposed Solutions


Detect Activity

Decode the data from the modulation alphabet A (e.g.,QPSK modulation)




Proposed Solutions


Detect ActivityDecode the data from the modulation alphabet A (e.g.,QPSK modulation)




Proposed Solutions

`1 norm recovery

MAP

x = minx∈A0

∥y −Hx∥22 + 2σ2

n∥x∥0 log((1 − pa)∣A∣pa

) (30)

MMSExMMSE = (HHH + σ2

nI)−1HHy (31)




Proposed Solutions

`1 norm recoveryMAP

x = minx∈A0

∥y −Hx∥22 + 2σ2


) (30)


nI)−1HHy (31)




Proposed Solutions

`1 norm recoveryMAP

x = minx∈A0

∥y −Hx∥22 + 2σ2


) (30)


nI)−1HHy (31)




Thank You!




Thank You!



Engineering

Interference management in spectrally and energy efficient wireless networks